Structural integrity, health and safety of hydraulic structures including dams.

我的研究领域深入到周围的结构健康和安全的水工建筑物，如大坝和堤坝的问题。这些结构提供了大量的好处;水坝的基本目的是促进水的储存。这些好处包括但不限于防洪、水力发电和人类消费。为了避免溃坝的灾难性后果，必须保证大坝结构的健康。我的研究重点将在于预测英国境内现实生活中的水工建筑物的失效机制。我想集中讨论的主要破坏机制是漫顶破坏，因为追溯到1799年至1925年，英国水坝协会将英国最常见的水坝破坏模式归因于洪水泛滥时的漫顶。环境署（2018年）在UKCP 18中报告说，由于气候变化，预测显示英国冬季降水量增加。一年中有些时候，由于降水量增加，大坝将储存更多的水，增加了漫顶的风险。随着时间的推移，降雨事件的趋势预计将进一步增加漫顶的风险。目前，洪水事件等自然灾害的预测准确性是工程界面临的主要挑战。一系列技术已被用来帮助预测洪水。其中一些技术包括Monte Carlo方法的变体或数字建模模拟，分别由Mohamed（2018）和Fenech等人（2019）进行。文献表明，洪水估计的预测方法的精度仍有很大的发展空间。近年来，被认为只有人类才能掌握的认知任务已经通过人工智能，即神经网络得到了发展。神经网络由算法设置，以检测和识别某些模式。Cichy和Kaiser在2019年的研究探索了基于大脑建模的强大算法，即深度神经网络（DNN）。DNN能够比其他模型更好地预测人脑的反应。这项研究的作者鼓励将DNN作为一个有用的模型，因为它们的预测能力很强大。Jin等人在2019年进一步支持了关于DNN的这一积极观点。对投影神经网络与机器人技术和模型预测控制的研究产生了有利于神经网络模型的结果，我的目标是将神经网络，以提高预测的准确性和精度增加的水位，以及在未来的水利结构内的任何其他异常。神经网络还将考虑所选水工结构的历史。因此，为了完成这项任务，我的目标是创建一个神经网络架构，以帮助时间序列预测和预报。由于目前的文献强烈支持使用CNN和LSTM的混合预测，我将使用这种架构类型来创建一个模型，可以帮助预测水工结构的失效机制。这将通过在水力结构上安装传感器来获得实时数据。这些传感器将安装在由环境署监管的真实大坝上。这将通过沟通和主管与机构和其他工业合作伙伴的联系来实现，例如拥有许多这些结构的Costain有限公司。这些数据将被输入深度学习模型进行训练，然后用于评估大坝的结构健康安全性，并预测其在极端天气事件下的性能。监测过程将在18个月期间进行，以获得一系列数据，说明季节性变化。

My research area delves into the issues surrounding the structural health and safety of hydraulic structures such as dams and levees. These structures provide a plethora of benefits; the fundamental purpose of a dam is to facilitate the storage of water. These benefits include, but are not limited to, flood control, hydropower, and human consumption. The structural health should be ensured to avoid the catastrophic event of dam failure, which can lead to severe consequences such as loss of property and even death. The focal point of my research will lie within predicting failure mechanisms of real-life hydraulic structures within the United Kingdom. The main failure mechanism I wish to focus on is breach by overtopping because dating back from 1799 to 1925, the British Dam Society attributes the most common modes of dam failure in the United Kingdom to overtopping during flooding. The Environment Agency (2018) reports in UKCP18 that the projections show an increase in winter precipitation over the United Kingdom, as a consequence of climate change. There will be times during the year where dams will be subjected to store more water due to the increased precipitation, increasing the risk of overtopping. As the years go on, the trend of rainfall events is expected to further add to the risk of overtopping.Currently, the accuracy in prediction of natural disasters such as flooding events present as a major challenge to the engineering world. A range of techniques have been used to aid in prediction of flooding. Some of these techniques included variants of the Monte Carlo method or digital modelling simulations, by Mohamed (2018) and Fenech et al (2019) respectively. Literature suggests there is still much room for development within the accuracy of prediction methods for flood estimation. Over recent years, cognitive tasks thought to be only mastered by humans have been developed through artificial intelligence, namely neural networks. Neural networks are set by algorithms to detect and recognise certain patterns. Cichy and Kaiser's study in 2019 explored the powerful algorithms that are deep neural networks (DNN), which are modelled based on the brain. It was found that DNNs were able to predict human brain responses better than other models. The authors of this study encourage embracing DNNs as a useful model as their capacity for prediction is powerful. This positive notion regarding DNNs is further supported by Jin et al. in 2019. The study on projection neural network with robotics and model predictive control yielded results that favoured the neural network model.I aim to incorporate neural networks to enhance the accuracy and precision of the prediction of increased water levels and as well as any other anomalies within the hydraulic structure in the future. The neural network would also consider the history of the chosen hydraulic structure. Therefore, to accomplish this task, I aim to create a neural network architecture to aid in time series prediction and forecasting. As current literature strongly supports the use of a hybrid of CNNs and LSTMs for forecasting, I would use this architecture type to create a model that can help predict failure mechanisms of hydraulic structures. This would be carried out by installing sensors on a hydraulic structure to obtain real-time data. The sensors will be installed on to a real-world dam regulated by the Environment Agency. This will be achieved through communications and the link the supervisor has with the agency and other industrial partners, such as Costain ltd who own many of these structures. The data will be fed into a deep learning model for training purposes which will then be used to evaluate the structural health safety of the dam and predict its performance under extreme weather events. The monitoring process would be carried out over an 18-month period to obtain a range of data to account for seasonal variation.